Events Calendar

CS Colloquium: George Porter: Towards Balanced, Data-intensive Scalable Computing

Speaker: George Porter, UC San Diego

Talk Title: Towards Balanced, Data-intensive Scalable Computing

Series: CS Colloquium

Abstract: While many interesting systems are able to scale linearly with additional servers, per-server performance can lag behind per-server capacity by more than an order of magnitude. In this talk, we will present Themis, a runtime supporting highly-efficient data-intensive computing. As an initial challenge application for this runtime, we built TritonSort, a highly efficient, scalable sorting system. It is designed to process large datasets with very high throughput (and has been evaluated against as much as 100 TB of input data spread across 832 disks in 52 nodes at a rate of 0.938 TB/min). It is also the winner of the 100TB "Indy" and "Daytona"
JouleSort benchmarks. In this talk, we will give an overview of the hardware and software architecture necessary to drive this level of efficiency. We then discuss how we have subsequetly generalized our system to support Map/Reduce programming. We believe the work holds a number of lessons for balanced system design and for scale-out architectures in general. Bridging the gap between high scalability and high performance will enable either significantly cheaper systems that are able to do the same work, or provide the ability to address significantly larger problem sets with the same infrastructure.

Biography: George Porter is a Research Scientist in the Center for Networked Systems and a member of the Systems and Networking group at UC San Diego. He received his B.S. from the University of Texas at Austin, and his Ph.D. from the University of California, Berkeley.

Host: Minlan Yu

Location: SSL 150

Audiences: Everyone Is Invited

Contact: Assistant to CS chair